Investors Shoot for the Moon

Quantinuum reported a logical error rate of about five errors per 1 million operations, roughly 10 times better than its record from a month earlier; Google said its magic-state cultivation process achieved 99.99% fidelity with about 40 times fewer errors, though both advances remain in testing and are not yet scaled for broad use. Deal Room figures cited on the podcast put deep-tech venture funding above $150 billion since the start of 2024, exceeding investment across the 2010s. The discussion also highlighted open-source AI as a potential threat to pricing power and returns on AI investment, while noting that cloud providers may still supply compute for those models.

Analysis

The investable distinction is between technical progress and monetizable advantage. Quantum milestones matter only if they translate into useful workloads at competitive cost; until then, QNT is exposed to financing, dilution, and valuation-duration risk. GOOG and IBM offer a less binary route because quantum upside sits alongside operating businesses, but it should not be capitalized as near-term earnings. Over the next 1–3 months, treat technical releases as sentiment catalysts, not proof of commercial readiness; over 6–18 months, watch independent replication, logical-qubit scaling, customer workloads, and funding needs.

For AI, open models are more likely to pressure model-layer pricing than eliminate compute demand. That supports cloud utilization in principle, but not automatically cloud returns: falling inference prices can outpace utilization gains, while heavy infrastructure commitments leave hyperscalers exposed to weaker returns on incremental capital. The second-order risk is to premium model providers and AI infrastructure suppliers whose economics depend on sustained scarcity or pricing power—not simply to cloud landlords. Power constraints may defer revenue conversion and favor grid, generation, and power-management capacity near term; fusion and other novel generation remain long-dated options, vulnerable to continued cost declines in renewables.

Contrarian point: reasonable current multiples do not protect earnings if AI prices reset, while technically impressive quantum headlines may be over-traded before commercial proof. No broad AI or quantum trade is warranted without entry valuation, capex-return, and balance-sheet data. Falsify the cautious AI view if cloud margins and returns on invested capital hold up as inference prices fall; falsify quantum optimism if milestones fail replication or scaling.

Key Decisions for Investors

• Prefer GOOG or IBM over a concentrated QNT position for quantum exposure; size QNT, if at all, as venture-like risk. Add only after independently verified scaling or commercial workload evidence, and reassess on financing/dilution disclosures.
• Monitor AMZN and MSFT cloud growth alongside segment margins, capex, and returns on incremental investment. Avoid buying utilization growth alone; weakening margins or rising capex without matching monetization would invalidate the landlord thesis.
• Keep AI model-layer exposure under review as open-source capability improves. Before taking a directional position in model providers or infrastructure names, verify pricing, customer retention, and workload mix; lower model prices with flat cloud margins would be a meaningful positive for cloud landlords.
• Treat novel power and eVTOL as watchlists rather than near-term trades. Favor evidence of signed, funded demand and deployable capacity over venture-funding totals; renewable cost declines or permitting delays could undermine longer-dated alternatives.